Introduction

Every year, the National Highway Traffic Safety Administration publishes the FARS, a census of all vehicle collision deaths on public roads. The basis for the FARS data is a partnership between the states and the National Highway Traffic Safety Administration to collect data on all qualifying fatalities. Death certificates are the source of race and ethnicity information. (Matthew & Ernani, 2022). For a collision to be covered by FARS, it must include a motor vehicle operating on a road that is normally accessible to the general public and must cause the death of one or more individuals—either non-motorists or occupants of the vehicle—within 30 days of the collision. In order to help identify traffic safety issues, provide an objective foundation for assessing the efficacy of motor vehicle safety regulations and highway safety initiatives, and provide an overall measure of highway safety, the U.S. Department of Transportation’s National Highway Traffic Safety Administration (DOT, NHTSA) developed the Framework for Assessment of Road Safety (FARS) in 1975 (DOT/NHTSA, n.d.).

This PROJECT analyzes data on traffic accidents in 2019, 2020, and 2021 to derive the relationship between traffic accidents and gender, age, address, time period and harm types.

This project includes the following:

  1. Data collection: Collect comprehensive data on traffic accidents for 2019, 2020 and 2021. This data should include details of the individuals involved (gender, age), location accident status (address), and time of occurrence.

  1. Data cleaning and preparation: standardize and clean the data to ensure consistency. This may involve dealing with missing values, correcting data entry errors, and formatting data for analysis.

  2. Exploratory data analysis (EDA): Descriptive analysis: generating statistics to understand the basic characteristics of the data, such as the total number of accidents per year, the distribution of accidents by gender and age, and the geographic distribution of accidents. Temporal analysis: examines trends over time, e.g., changes in accident frequency over a three-year period or over a specific time period, e.g., daily peaks, daytime versus nighttime.

  3. Correlation analysis:

    Gender analysis: investigates how gender affects the likelihood or severity of traffic crashes.

    Age analysis: to determine whether certain age groups are more likely to be involved in traffic accidents.

    Geospatial analysis: analyzes whether certain addresses or areas (this PROJECT focuses specifically on New Jersey and compares it to the U.S. as a whole) have higher accident rates, potentially identifying high-risk areas.

    Temporal analysis: exploring the relationship between different times of the day (hours of the day, months of the year (this project differentiates the year with 4 quarters)) and accident rates.

    Type analysis: Analysis of the frequency of the types of accidents, the difference between the types with higher frequency of accidents in the three years.

  4. Insights and recommendations: Derive insights from the analysis. For example, determine if certain demographics are at greater risk or if certain times or locations are associated with higher crash rates. Based on these findings, develop recommendations for traffic safety improvements or targeted campaigns.

  5. Reporting and Visualization: Detailed descriptions of the findings are accompanied by visualizations such as charts, heat maps, and graphs to make the data easy to understand.

Boxplot Diagram

Boxplot Diagram

This is a picture of New Year’s Day car crash in NYC leaves couple dead,their car rammed by a pickup truck whose driver then bolted on foot.

Load data

Import packages

library(dplyr) # this package is used for data manipulation
library(knitr)  # for table formatting
library(kableExtra) # more advanced formatting options in tables
library(tidyr) # providing a set of functions for transforming the layout of data sets
library(ggplot2) # for data visualization
library(tidyverse) # providing consistent data structures
library(patchwork) # combine multiple ggplot objects

Reading and cleaning the data, prepare data to analysis

Clean and select accident data

# read files accident form 2021, 2020 and 2019 and combine them in one file call that accident_all
accident_2019 <- read.csv("accident_2019.csv")
accident_2020 <- read.csv("accident_2020.csv")
accident_2021 <- read.csv("accident_2021.csv")

accident_2019[] <- lapply(accident_2019, as.character)
accident_2020[] <- lapply(accident_2020, as.character)
accident_2021[] <- lapply(accident_2021, as.character)

accident_2019$YEAR <- 2019
accident_2020$YEAR <- 2020
accident_2021$YEAR <- 2021

# merge data
accident_all <- bind_rows(accident_2019, accident_2020, accident_2021)

# data cleaning: remove missing information from data to reduce interference with data analysis
accident_all <- accident_all %>% filter(complete.cases(.) & !apply(., 1, function(x) any(x == "")))

# select data: This project focuses on the trend of traffic harm types of accident over 3 years, so HARM_EVNAME and YEAR selected.
accident_all <- accident_all %>%
  select(HARM_EVNAME,YEAR)

head(accident_all)
##                      HARM_EVNAME YEAR
## 1 Immersion or Partial Immersion 2019
## 2     Motor Vehicle In-Transport 2019
## 3     Motor Vehicle In-Transport 2019
## 4     Motor Vehicle In-Transport 2019
## 5       Other Object (not fixed) 2019
## 6     Motor Vehicle In-Transport 2019

Clean and select person data

# read files accident form 2021, 2020 and 2019 and combine them in one file call that accident_all
person_2019 <- read.csv("person_2019.csv")
person_2020 <- read.csv("person_2020.csv")
person_2021 <- read.csv("person_2021.csv")

person_2019[] <- lapply(person_2019, as.character)
person_2020[] <- lapply(person_2020, as.character)
person_2021[] <- lapply(person_2021, as.character)

person_2019$YEAR <- 2019
person_2020$YEAR <- 2020
person_2021$YEAR <- 2021

# merge data
person_all <- bind_rows(person_2019, person_2020, person_2021)

# data cleaning: remove missing information from data to reduce interference with data analysis
person_all <- person_all %>% 
              # Remove rows with any empty strings in key columns
              filter(!apply(select(., AGE, SEXNAME, STATENAME, YEAR, MONTH,HOUR, DEATH_YR), 1, function(x) any(x == ""))) %>%
              # Convert AGE to numeric 
              mutate(AGE = as.numeric(AGE)) %>%
              # Remove rows where AGE data is wrong which is greater than 900
              filter(AGE <= 900)


# select data: This project focuses on the relationship between traffic accidents and gender, age, address, and time of accident, so AGE,SEXNAME,STATENAME,YEAR,MONTH,DEATH_YR were selected.
person_all <- person_all %>%
  select(AGE,SEXNAME,STATENAME,YEAR,MONTH,HOUR,DEATH_YR)

head(person_all)
##   AGE SEXNAME STATENAME YEAR MONTH HOUR DEATH_YR
## 1  34  Female   Alabama 2019     2   12     8888
## 2  53    Male   Alabama 2019     2   12     2019
## 3  59    Male   Alabama 2019     2   12     8888
## 4  42  Female   Alabama 2019     1   18     2019
## 5  54  Female   Alabama 2019     1   18     8888
## 6  22    Male   Alabama 2019     1   19     8888

The relationship between gender and traffic accidents

Compare the proportion of men and women involved in accidents

# Grouping by Gender and count how many men and women 
table(person_all$SEXNAME)
## 
##              Female                Male        Not Reported Reported as Unknown 
##               84526              173978                 392                  66
person_all<-person_all|> 
  mutate(gender= case_when(
SEXNAME=="Female"~"Female",
SEXNAME=="Male"~"Male",
TRUE~"Unknown"))
total_gender <- person_all %>%
  group_by(SEXNAME) %>%
  summarise(Total = n()) %>%
  ungroup()

gender_with_percentages <- person_all %>%
                          group_by(gender) %>%
                          summarise(Count = n()) %>%
                          mutate(Percentage = (Count / sum(Count)) * 100)

# Tidy up the table
gender_with_percentages <- gender_with_percentages %>%
                          mutate(Percentage = round(Percentage, 2))

# print the result
kable(gender_with_percentages)
gender Count Percentage
Female 84526 32.64
Male 173978 67.18
Unknown 458 0.18

  1. There are a total of 84,526 cases involving females, which accounts for 32.64% of the dataset.

  2. There are a total of 173,978 cases involving males, which accounts for 67.18% of the dataset.

    From this data, we can see that a significantly larger portion of the dataset consists of males. Males are more frequently involved in accidents than females.

Compare the percentage of Survived between male and female

# Count the number and proportion of Survived for each gender
person_all<-person_all|> mutate(Death= 
                case_when(
DEATH_YR%in%c(2019,2020,2021,2022)~"Died",
DEATH_YR==8888~"Survived", TRUE~"Unknown"))

# Reshaping the data
survived_gender_wide <- person_all %>%
  filter(Death == "Survived", SEXNAME %in% c("Female", "Male")) %>%
  group_by(YEAR, SEXNAME) %>%
  summarise(Count = n(), .groups = 'drop') %>%
  mutate(Percent = round((Count / sum(Count)) * 100, 1)) %>%
  pivot_wider(names_from = SEXNAME, values_from = c(Count, Percent), 
              names_sep = "_", values_fill = list(Count = 0, Percent = 0))

# Format the table using kable
kable(survived_gender_wide, col.names = c("Year", "Survived Male Count", "Survived Female Count", "Survived Male Percent", "Survived Female Percent"))
Year Survived Male Count Survived Female Count Survived Male Percent Survived Female Percent
2019 16552 28112 11.8 20.0
2020 16063 28986 11.4 20.6
2021 18711 32425 13.3 23.0

Women have a higher survival rate than men.

Comparing data for the US and NJ (age)

Compare the average age of people of New Jersey accident data to the US data

# Prepare the New Jersey dataset
person_nj <- filter(person_all, STATENAME == "New Jersey")
person_nj$Group <- "New Jersey"

# Prepare the overall dataset with a Group column
person_all$Group <- "US"

# Combine the data
combined_data <- rbind(person_all[, c("AGE", "Group")], person_nj[, c("AGE", "Group")])

# Calculate mean ages
mean_ages <- combined_data %>%
             group_by(Group) %>%
             summarise(MeanAge = mean(AGE, na.rm = TRUE))

# Create the boxplot with mean age highlighted
ggplot(combined_data, aes(x = Group, y = AGE, fill = Group)) +
  geom_boxplot() +
  geom_point(data = mean_ages, aes(x = Group, y = MeanAge, group = Group), 
             color = "red", size = 3, shape = 18) +
  geom_text(data = mean_ages, aes(x = Group, y = MeanAge, label = round(MeanAge, 1)), 
            vjust = -1, color = "red") +
  labs(title = "Boxplot of Ages: US vs. New Jersey", x = "Group", y = "Age") +
  theme_minimal()

  1. The horizontal line within each box represents the median age for both groups: person_nj (New Jersey) and person_all (United States). The relatively similar medians indicate that the average age of individuals involved in accidents in New Jersey and the United States as a whole is almost the same.

  2. The center 50% of the data are included in the interquartile range (IQR), which is shown by the boxes. The equal magnitude of the IQRs for the two groups suggests that the middle 50% of cases in both datasets have a similar age distribution.

  3. There appears to be little variation in the general age distribution based on the “whiskers” of the boxplot, which show the range of the data excluding outliers.

  4. There do not appear to be any outliers indicated, which would show data points that fall significantly outside the typical range.

Overall, the average age of people involved in accidents in New Jersey does not differ significantly from the national average.

Compare the age group with the highest concentration of accidents in New Jersey and the United States

# Function to categorize age into segments
categorize_age <- function(age) {
  cut(age,
      breaks = seq(0, 100, by = 10),
      include.lowest = TRUE,
      right = FALSE,
      labels = paste(seq(0, 90, by = 10), seq(9, 99, by = 10), sep = "-"))
}

# Categorize age into segments for both datasets
person_all <- person_all %>%
  mutate(AGE_GROUP = categorize_age(AGE))

person_nj <- person_nj %>%
  mutate(AGE_GROUP = categorize_age(AGE))

# Calculate the percentage of each age group for the US
age_group_percentage_us <- person_all %>%
  count(AGE_GROUP) %>%
  mutate(Percentage = n / sum(n) * 100) %>%
  arrange(desc(Percentage))

# Calculate the percentage of each age group for NJ
age_group_percentage_nj <- person_nj %>%
  count(AGE_GROUP) %>%
  mutate(Percentage = n / sum(n) * 100) %>%
  arrange(desc(Percentage))

# Find the age group with the highest concentration of accidents for US and NJ
top_age_group_us <- age_group_percentage_us[1, ]
top_age_group_nj <- age_group_percentage_nj[1, ]

# Compare the top age groups
comparison <- rbind(
  data.frame(Location = "US", top_age_group_us),
  data.frame(Location = "NJ", top_age_group_nj)
)

# Print the comparison
kable(comparison)
Location AGE_GROUP n Percentage
US 20-29 56551 21.83757
NJ 20-29 884 21.13316

The age group with the highest concentration of accidents in New Jersey and the United States is 20 to 29 years old, and the proportion is close.

  1. High-risk age group: The 20-29 age group is usually associated with young people. This group may exhibit a higher level of risk-taking behavior, which may lead to a higher rate of accidents. This is a trend in New Jersey and across the United States.

  2. Consistency across regions: New Jersey has similar rates to the rest of the U.S., which means that the factors contributing to this trend are not unique to New Jersey, but are likely to be prevalent….

  3. Research on influencing factors: Further research into why the 20-29 age group is accident prone may be helpful. Factors may include inexperience, overconfidence, higher rates of drunk driving or greater exposure to high-risk driving situations.

The association of the time period and traffic accident

Count the 4-hour periods in which traffic accidents occur most frequently

# Function to map hours to the four-hour intervals
map_to_interval <- function(hour) {
  # Extract the starting hour from the HOURNAME string
  start_hour <- as.numeric(str_extract(hour, "^\\d+"))
  
  # Define the intervals
  if (start_hour >= 0 & start_hour < 4) {
    return("0:00am-3:59am")
  } else if (start_hour >= 4 & start_hour < 8) {
    return("4:00am-7:59am")
  } else if (start_hour >= 8 & start_hour < 12) {
    return("8:00am-11:59am")
  } else if (start_hour >= 12 & start_hour < 16) {
    return("12:00pm-3:59pm")
  } else if (start_hour >= 16 & start_hour < 20) {
    return("4:00pm-7:59pm")
  } else if (start_hour >= 20 & start_hour < 24) {
    return("8:00pm-11:59pm")
  } else {
    return("Unknown")
  }
}

# Apply the function to group 'HOURNAME' into four-hour intervals
person_all <- person_all %>%
  mutate(Interval = sapply(HOUR, map_to_interval))

# Count the occurrences of each interval and calculate the percentage
interval_counts <- person_all %>%
  count(Interval, name = "Count") %>%
  mutate(Percentage = Count / sum(Count) * 100) %>%
  arrange(desc(Count))

# Tidy the table
interval_counts <- interval_counts %>%
                          mutate(Percentage = round(Percentage, 2))

# View the results
kable(interval_counts)
Interval Count Percentage
4:00pm-7:59pm 61474 23.74
8:00pm-11:59pm 55344 21.37
12:00pm-3:59pm 49242 19.02
0:00am-3:59am 31693 12.24
8:00am-11:59am 30530 11.79
4:00am-7:59am 29529 11.40
Unknown 1150 0.44 
# Create a pie chart with percentage labels and interval names
pie_chart <- ggplot(interval_counts, aes(x = "", y = Percentage, fill = Interval)) +
  geom_bar(width = 1, stat = "identity") +
  coord_polar("y", start = 0) +
  geom_text(aes(label = paste0(Interval, "\n", round(Percentage, 1), "%")), 
            position = position_stack(vjust = 0.5)) +
  theme_void() +
  labs(fill = "Interval", title = "Frequency of accidents during 4-hour periods") +
  theme(legend.position = "bottom")


# Print the pie chart
print(pie_chart)

  1. Peak Hours: The intervals from 4:00 pm to 7:59 pm and from 8:00 pm to 11:59 pm have the highest counts, indicating that these time periods are peak hours for the accidents occured. This could suggest higher activity levels or increased risk during these times.

  2. Day vs. Night: There is a noticeable decrease in counts during the late night to early morning hours (0:00 am to 3:59 am), which may indicate lower activity levels or that fewer accidents are occured during these hours.

  3. Afternoon Activity: The interval from 12:00 pm to 3:59 pm also has a significant count, which could correspond to afternoon activities or occurrences.

  4. Morning Hours: The intervals from 8:00 am to 11:59 am and from 4:00 am to 7:59 am have lower counts compared to the afternoon and evening hours. This may reflect a slower start to the day for the events being recorded or may indicate that the risk or activity is lower during these times.

Frequency of car accidents per quarter

# Create a new variable 'Quarter' by mapping 'MONTH' to quarters
person_all <- person_all %>%
  mutate(Quarter = case_when(
    MONTH %in% 1:3 ~ "Quarter_1",
    MONTH %in% 4:6 ~ "Quarter_2",
    MONTH %in% 7:9 ~ "Quarter_3",
    MONTH %in% 10:12 ~ "Quarter_4",
    TRUE ~ "Unknown"  # For any data that might not fit the above criteria
  ))

# Count the occurrences of each quarter and calculate the percentage
quarter_counts <- person_all %>%
  count(Quarter, name = "Count") %>%
  mutate(Percentage = Count / sum(Count) * 100) %>%
  arrange(desc(Count))

# Tidy the table
quarter_counts <- quarter_counts %>%
                          mutate(Percentage = round(Percentage, 2))

# View the results
kable(quarter_counts)
Quarter Count Percentage
Quarter_3 72247 27.90
Quarter_4 68554 26.47
Quarter_2 64027 24.72
Quarter_1 54134 20.90 
# Create a pie chart with labels for each section
ggplot(quarter_counts, aes(x = "", y = Percentage, fill = Quarter)) +
  geom_bar(width = 1, stat = "identity") +
  coord_polar("y", start = 0) +
  geom_text(aes(label = sprintf("%s: %.1f%%", Quarter, Percentage)), 
            position = position_stack(vjust = 0.3)) +
  theme_void() +
  labs(fill = "Quarter", title = "Frequency of accidents during quarters") +
  theme(legend.position = "bottom")

  1. Quarter 3 (Q3) has the highest count: With 72,247 occurrences, making up 27.90% of the total, this suggests that the events being recorded happen most frequently during the months of July, August, and September.

  2. Quarter 4 (Q4) follows closely: With 68,554 occurrences and 26.47% of the total, indicating that October, November, and December are also high-activity months.

  3. Quarter 2 (Q2) is the next: Having 64,027 occurrences with 24.72% of the total, which includes the months of April, May, and June.

  4. Quarter 1 (Q1) has the fewest occurrences: With 54,134 occurrences accounting for 20.90% of the total, which suggests that January, February, and March are the least active months.

It might indicate seasonal patterns in driving behavior or road conditions that could lead to higher incidents in Q3 and Q4.

Table of frequency of occurrence of the time period in each quarter

# Calculate the total number of accidents for each quarter
total_accidents_per_quarter <- person_all %>%
  group_by(Quarter) %>%
  summarise(Total_Count = n(), .groups = "drop")

# Now, for each quarter, find the 4-hour period with the highest frequency of accidents
most_frequent_intervals <- person_all %>%
  group_by(Quarter, Interval) %>%
  summarise(Count = n(), .groups = "drop") %>%
  arrange(Quarter, desc(Count))

# Merge the total counts back into the most frequent intervals
most_frequent_intervals <- most_frequent_intervals %>%
  left_join(total_accidents_per_quarter, by = "Quarter")

# Calculate the percentage
most_frequent_intervals <- most_frequent_intervals %>%
  mutate(Percentage = (Count / Total_Count) * 100)

# Get the top interval for each quarter with the percentage
top_intervals_by_quarter <- most_frequent_intervals %>%
  group_by(Quarter) %>%
  slice_max(n = 1, order_by = Count) %>%
  ungroup()

# Tidy the table
top_intervals_by_quarter <- top_intervals_by_quarter %>%
                          mutate(Percentage = round(Percentage, 2))

# Print the results
kable(top_intervals_by_quarter)
Quarter Interval Count Total_Count Percentage
Quarter_1 4:00pm-7:59pm 13779 54134 25.45
Quarter_2 8:00pm-11:59pm 14496 64027 22.64
Quarter_3 8:00pm-11:59pm 16799 72247 23.25
Quarter_4 4:00pm-7:59pm 18893 68554 27.56

Analysis based on the information of the table:

  1. Quarter 1 (Q1): The 4:00 pm to 7:59 pm interval had the highest frequency of car accidents, with 13,779 accidents, constituting 25.45% of the total accidents in Q1.

  2. Quarter 2 (Q2): The interval with the highest number of accidents was from 8:00 pm to 11:59 pm, with 14,496 accidents, which is 22.64% of Q2’s total.

  3. Quarter 3 (Q3): This quarter also saw the highest number of accidents in the 8:00 pm to 11:59 pm interval, with 16,799 accidents, making up 23.25% of the total for Q3.

  4. Quarter 4 (Q4): The 4:00 pm to 7:59 pm interval was the most accident-prone, with 18,893 accidents, representing 27.56% of Q4’s total accidents.

Analysis reveals several patterns and potential insights:

  1. The intervals of 4:00 pm to 7:59 pm and 8:00 pm to 11:59 pm are consistently the most accident-prone across all quarters, suggesting that the later afternoon to late evening may be the riskiest times for car accidents.
  2. Quarter 4 shows the highest percentage (27.55%) of accidents in the peak interval compared to other quarters, which could be influenced by factors such as early darkness in the winter months, holiday traffic, or other seasonal factors.
  3. The fact that the 8:00 pm to 11:59 pm interval is the peak in Q2 and Q3 could suggest that summer evening activities contribute to the higher accident rates during these hours.
  4. Quarter 1 has the lowest percentage of accidents in its peak interval, indicating perhaps more evenly distributed accident occurrences throughout the day or less pronounced peaks compared to other quarters.

Conclusion

  1. Gender differences in accidents: Males accounted for a larger proportion of accidents at 65.84%, followed by females at 32.01% and unknown gender at 2.15%, with females having a higher survival rate in accidents.

  2. New Jersey and the U.S. have the highest concentration of accidents in the 20-29 age group, and the average age of those involved in accidents is similar. Suggesting that New Jersey has similar rates to the rest of the U.S. means that contributing to this trend is likely to be widespread.

  3. Peak Accident Hours:Peak accident hours are 4:00 p.m. to 7:59 p.m. and 8:00 p.m. to 11:59 p.m., which indicate higher activity levels or higher risk. Counts were reduced during the day compared to the night, while activity between 12:00 PM and 3:59 PM was significant. Lower counts in the morning may reflect slower starts or lower risk.

  4. Seasonal variation in traffic accidents: The data show seasonal trends, with the highest frequency of traffic accidents occurring in the third quarter (July-September) and the fourth quarter (October-December). This can be attributed to increased summer and holiday travel, as well as poor weather conditions affecting driving safety in the fourth quarter. Enhanced traffic safety measures and controls are needed. Initiatives may include increased enforcement efforts and public safety activities designed to anticipate and mitigate increased risk.

  5. Annual accident trend: Persistent high rates of certain types of crashes over the years indicate underlying systemic problems. Assessing crash frequency and type can assist the transportation sector in developing and implementing strategic interventions to minimize crashes. For example, addressing the issue of “rollover/flip”. Accidents may require a review of vehicle safety standards and road design.

  6. Informed decision-making through data visualization Clear visualization of accident data helps decision-makers to quickly identify and prioritize safety issues, underscoring the importance of reliable data collection and representation in policy development. Gaining insight into the timing and nature of crashes and understanding crash trends can help to take proactive measures to anticipate potential future crash hotspots and periods of elevated risk, and predictive analytics can help to better allocate resources and emergency response plans, which can ensure preparedness during periods of high risk.